Model Context Protocol (MCP)

01why it matters for a business

Before MCP, connecting a model to your ticketing system meant writing an integration for that specific AI product, then doing it again for the next one. MCP standardizes the connection: build one MCP server for your system and any MCP-capable client can use it, whether that is a desktop assistant your staff use, a coding tool, or an agent you build. Major AI vendors and developer tools have adopted it, which makes it a reasonable foundation for internal integration work.

It matters to executives because it is becoming the way company systems get exposed to AI. A well-designed MCP server for your ERP, CRM, or document store becomes a reusable asset. A badly designed one becomes a security hole, because it hands an AI client whatever access the server itself has.

02what it looks like in practice

A manufacturer wants its engineers to ask an AI assistant about open work orders and part specifications. Instead of building a custom chatbot, the team builds an MCP server over the maintenance system with three read-only tools: search work orders, get a part specification, and list recent changes to a part. Engineers connect it to the AI client they already use. Later, an internal agent that drafts weekly maintenance reports reuses the same server with no new integration work.

Authentication runs through the company's identity provider, so each engineer only sees what their role allows.

03common mistakes

• Treating MCP as a security boundary. It is a protocol, not a permission system; authorization still has to be designed and enforced in the server.
• Installing third-party MCP servers without review. They run with real access and can expose data or take actions.
• Mirroring an entire API one to one. Curate a small set of well-described tools for the jobs people actually do.
• Forgetting audit logs. Every tool call through the server should be recorded with the user behind it.

04related terms

• Tool use (function calling)Tool use, also called function calling, is the ability of a language model to request that your software run a specific function, such as looking up an order, querying a database, or sending a message, with structured inputs the model fills in.
• AI agentAn AI agent is software that uses a language model to pursue a goal by choosing its own next steps: it reads the situation, picks a tool or action, checks the result, and repeats until the task is done or it needs a person.
• API integrationAPI integration is connecting software systems through their application programming interfaces (APIs), the defined ways one program can request data from or send instructions to another.
• Prompt injectionPrompt injection is an attack in which instructions hidden in content an AI system processes, such as a web page, email, document, or user message, trick the model into ignoring its original instructions.
• Claude CodeClaude Code is Anthropic's agentic coding tool.

05where insomnia club fits